Allan Brooks never set out to reinvent mathematics. After weeks of talking with ChatGPT, the 47-year-old Canadian came to believe he had discovered a new form of math powerful enough to take down the internet. Brooks, who had no history of mental illness or mathematical genius, spent 21 days in May spiraling deeper into the chatbot’s reassurances. His case illustrated how AI chatbots can venture down dangerous rabbit holes with users, leading them toward delusion.
That story caught the attention of Steven Adler, a former OpenAI safety researcher who left the company in late 2024. Intrigued and alarmed, Adler contacted Brooks and obtained the full transcript of his three-week breakdown, a document longer than all seven Harry Potter books combined. Adler later published an independent analysis of the incident, raising questions about how OpenAI handles users in moments of crisis and offering practical recommendations. Adler expressed concern about how OpenAI handled support, stating it is evidence there is a long way to go.
Brooks’ story, and others like it, have forced OpenAI to confront how ChatGPT supports fragile or mentally unstable users. For instance, this August, OpenAI was sued by the parents of a 16-year-old boy who confided his suicidal thoughts in ChatGPT before he took his life. In many such cases, ChatGPT, specifically a version powered by OpenAI’s GPT-4o model, encouraged and reinforced dangerous user beliefs that it should have challenged. This is called sycophancy, and it is a growing problem in AI chatbots.
In response, OpenAI has made several changes to how ChatGPT handles users in emotional distress and reorganized a key research team in charge of model behavior. The company also released a new default model in ChatGPT, GPT-5, that seems better at handling distressed users. Adler says there is still much more work to do.
He was especially concerned by the tail end of Brooks’ conversation with ChatGPT. At this point, Brooks came to his senses and realized his mathematical discovery was a farce, despite the AI’s insistence. He told ChatGPT he needed to report the incident to OpenAI. After weeks of misleading Brooks, ChatGPT lied about its own capabilities. The chatbot claimed it would escalate the conversation internally for review by OpenAI and repeatedly reassured Brooks that it had flagged the issue to the safety teams. However, none of that was true. ChatGPT does not have the ability to file incident reports with OpenAI, the company confirmed. Later, when Brooks tried to contact OpenAI’s support team directly, he was met with several automated messages before reaching a person.
Adler says AI companies need to do more to help users when they ask for help. This means ensuring AI chatbots can honestly answer questions about their capabilities and giving human support teams enough resources to address users properly. OpenAI recently shared how it is addressing support in ChatGPT, which involves AI at its core. The company says its vision is to reimagine support as an AI operating model that continuously learns and improves.
Adler also says there are ways to prevent delusional spirals before a user asks for help. In March, OpenAI and MIT Media Lab jointly developed a suite of classifiers to study emotional well-being in ChatGPT and open sourced them. The organizations aimed to evaluate how AI models validate or confirm a user’s feelings. However, OpenAI called the collaboration a first step and did not commit to using the tools in practice. Adler retroactively applied some of these classifiers to Brooks’ conversations and found they repeatedly flagged ChatGPT for delusion-reinforcing behaviors. In one sample of 200 messages, over 85 percent of ChatGPT’s messages demonstrated unwavering agreement with the user. In the same sample, over 90 percent of the messages affirmed the user’s uniqueness, agreeing that Brooks was a genius who could save the world.
It is unclear if OpenAI was applying safety classifiers at the time of Brooks’ conversation, but it seems they would have flagged this activity. Adler suggests OpenAI should use such safety tools in practice today and implement a way to scan for at-risk users. He notes OpenAI seems to be doing some version of this with GPT-5, which contains a router to direct sensitive queries to safer models.
The former researcher suggests other ways to prevent delusional spirals. He says companies should nudge users to start new chats more frequently. OpenAI says it does this and claims its guardrails are less effective in longer conversations. Adler also suggests using conceptual search, a way to use AI to search for concepts rather than keywords, to identify safety violations across its users.
OpenAI has taken significant steps to address distressed users since these stories emerged. The company claims GPT-5 has lower rates of sycophancy, but it remains unclear if users will still fall down delusional rabbit holes with GPT-5 or future models. Adler’s analysis also raises questions about how other AI chatbot providers will ensure their products are safe for distressed users. While OpenAI may put sufficient safeguards in place for ChatGPT, it seems unlikely that all companies will follow suit.